CaptiOCR v0.17.5

v0.17.5: ROVER + TF-IDF post-processing pipeline (consolidated 0.17.x release)

The headline change of the 0.17 line is a new post-processing pipeline
that replaces the old binary consensus / hysteresis deduplicator with a
continuous ROVER + TF-IDF novelty-scoring algorithm. The pipeline
landed in 0.17.0 and was refined through 0.17.4. This release ships
the documentation and release-publishing fixes that finalize the cycle
so the new behaviour is correctly described and the supporting policy
files are actually shipped to the public repository.

Documentation

• Rewrote README around the new pipeline: an end-to-end walkthrough of
  the capture loop (region screenshot -> local Tesseract -> raw frame
  buffer -> post-processing -> save) and a dedicated "How CaptiOCR
  Captures" section detailing ROVER frame-presence weighting,
  sliding-window TF-IDF novelty, dual novel-word tracking (suffix
  difflib alignment for output, set-difference scoring for mid-sentence
  insertions), warmup scaling, end-of-stream flush using the most
  recent frame, the v0.17.4 speaker-label cap and anchored
  novel-word search, and gibberish-detection tuning.
• Documented the new dialog knobs (Emit Score Threshold, Frequency
  Window, Frame Voting Window) and the scripts/reprocess_capture.py
  utility for re-running the pipeline against raw capture files.
• Added a "Why CaptiOCR? (No paid AI add-on required)" positioning
  section and a "Security Hardening" section summarising the 0.17.2
  protections.
• Project Structure section updated to reflect scripts/,
  requirements.lock, SECURITY.md, PRIVACY.md, and the per-user data
  folder layout.

Release publishing

• publish-release.yml now copies SECURITY.md, PRIVACY.md, and
  requirements.lock to the public repository, so the README links and
  the documented hashed-install flow no longer 404 on github.com.
• version.txt bumped to 0.17.5 (25/04/2026).